WeatherRAN Dashboard

Weather-predictive O-RAN xApp for rural Canadian 5G networks | GitHub

5.3% Signal Improvement

Fewer errors during rainstorms by predicting weather and adapting the radio before signal degrades

What does this mean in plain language?
When it rains, cell phone signals get worse — calls drop, video buffers, IoT sensors lose connection. Traditional cell towers wait until the signal is already bad, then try to fix it. WeatherRAN reads the weather forecast and adjusts the radio settings before the rain even hits. Result: the signal stays stronger through the storm.

Key Results

5.3%

Prairie improvement (N=1000)

3.3%

Forest improvement (N=1000)

Canadian terrains modelled

332

Tests passing

All 4 Canadian Terrains (N=1,000 each)

Saskatchewan Prairie

Flat farmland — rain is the main challenge

Error rate without	13.02%
Error rate with WeatherRAN	12.33%
Signal loss	107.7 dB

Improved:

5.3%

Ontario Boreal Forest

Dense trees block signal — rain adds to foliage loss

Error rate without	15.69%
Error rate with WeatherRAN	15.17%
Signal loss	118.1 dB

Improved:

3.3%

BC Rocky Mountains

Mountains block and scatter signal — valleys trap echoes

Error rate without	14.22%
Error rate with WeatherRAN	13.85%
Signal loss	118.5 dB

Improved:

2.5%

Arctic Tundra

Frozen ground reflects signal — ice coats antennas at -30C

Error rate without	12.99%
Error rate with WeatherRAN	12.99%
Signal loss	109.2 dB

Improved:

0.0%

Why do results vary by terrain?
Prairie — rain is the main problem on flat land, so weather prediction helps most (5.3%).
Forest — trees already block signal; rain adds less relative damage (3.3%).
Mountains — ridges and valleys scatter signal; rain on top is secondary (2.6%).
Arctic — at -30C with light precipitation, rain threshold isn't met. Arctic challenges (ice loading, permafrost reflection, blizzard scattering) are handled by the terrain model itself. The TN/LEO failover engine provides the real value here — keeping connectivity when conditions are extreme.

6 Canadian Cell Sites

Map shows live Environment Canada weather radar (MSC GeoMet WMS — anonymous, no API key). Click a site for details. Run full map locally: uvicorn src.ran_intel.app:app --port 8080

How It Works (Simple Version)

#	Step	What happens
1	Read weather	Free Government of Canada weather data (no API key needed) tells us rain is coming
2	Predict impact	Our terrain model calculates how much the rain will weaken the signal for this specific landscape
3	Adapt radio	Before rain arrives: switch to a more robust radio mode (like speaking slower on a noisy phone call)
4	Rain hits	Signal stays strong because we prepared. No dropped calls, no buffering.

The old way vs our way:
Old way: Rain hits → signal drops → tower detects problem → fixes it → user already had a bad experience.
Our way: Forecast says rain coming → fix settings now → rain hits → user never notices.

Software Quality

ALL GREEN — 332/332 tests passed

Every component is tested. Every test passes. Zero failures.

Module	Tests
adapters/test_iot_ingestion.py	✓ 20/20
adapters/test_weather_gc_adapter.py	✓ 13/13
channel/arctic_tundra/test_scene.py	✓ 14/14
channel/boreal_forest/test_scene.py	✓ 15/15
channel/prairie_rma/test_scene.py	✓ 16/16
channel/rocky_mountain/test_scene.py	✓ 16/16
integration/test_e2e_pipeline.py	✓ 11/11
integration/test_latency_budget.py	✓ 5/5
policies/test_beam_adaptation_policy.py	✓ 28/28
policies/test_dnd_priority_queue.py	✓ 7/7
policies/test_iot_priority_scheduler.py	✓ 28/28
policies/test_ntn_handover_predictor.py	✓ 25/25
policies/test_spectrum_anomaly_policy.py	✓ 29/29
policies/test_tn_leo_failover.py	✓ 23/23
policies/test_weather_mcs_policy.py	✓ 16/16
smoke/test_budget_check.py	✓ 5/5
smoke/test_demo_app.py	✓ 25/25
smoke/test_integration_benchmark.py	✓ 5/5
smoke/test_lineage_audit.py	✓ 5/5
smoke/test_ntn_coverage_freshness.py	✓ 5/5
smoke/test_protected_b.py	✓ 5/5
smoke/test_ran_intel.py	✓ 6/6
smoke/test_spec_version_check.py	✓ 3/3
smoke/test_stride_check.py	✓ 3/3
smoke/test_weather_api_check.py	✓ 4/4

All Deliverables (Phase 1 + 2 + 3)

Deliverable	Phase	What it means
Done Weather adapter	1	Reads live Canadian weather data — free, no API key
Done Prairie channel	1	Saskatchewan flat farmland — 3GPP RMa + rain attenuation
Done Boreal forest channel	1	Ontario/Quebec forests — foliage blockage + snow
Done MCS adjustment policy	1	Rain detected → radio switches to robust mode automatically
Done Beam adaptation policy	1	Widens antenna beam during storms for wider coverage
Done Docker demo	1	`docker compose up` — runs entire demo on any laptop
Done Rocky mountain channel	2	BC/Alberta — mountain diffraction + valley multipath
Done Arctic tundra channel	2	Northern Canada — permafrost + ice loading + blizzard
Done RAN-Intel live map	2	6 Canadian sites with live weather radar (see map above)
Done Anomaly detection	2	Detects interference, DoS, signal manipulation
Done NTN handover predictor	2	Predicts satellite dropout 60s ahead (F1 >= 0.80)
Done STRIDE security models	2	Threat models for all API boundaries
Done IoT priority scheduler	3	URLLC/eMBB/mMTC scheduling — pipelines and rail get priority
Done TN/LEO failover	3	Auto terrestrial-satellite switching with anti-flapping
Done IoT ingestion layer	3	MQTT/AMQP message routing — defence devices to secure queue
Done DND priority queue	3	PROTECTED-B encryption + human approval enforced
Done PROTECTED-B compliance	3	Data sovereignty checks — classified data never leaves Canada
Done Per-hop latency tests	3	Uu ≤3ms, RAN ≤2ms, backhaul ≤3ms, app ≤2ms, E2E ≤10ms
Done CI pipeline	2	5-job GitHub Actions: monitor, smoke, integration, security, benchmark
Done Jetson Orin deploy	2	ARM64 edge deployment for field use

For Engineers: Architecture

api.weather.gc.ca ──GET──▶ Weather Adapter ──▶ Policy Engine ──▶ E2SM-RC Control ──▶ gNB
(anonymous, free)          (retry, log)        6 policies:        (O-RAN E2)
                                               ├ WeatherMCS       Applied at next
                           Channel Models      ├ BeamAdapt        scheduling slot
                           ├ Prairie (3GPP)    ├ Anomaly          (< 10ms E2E)
                           ├ Boreal (foliage)  ├ NTN Handover
                           ├ Mountain (diffr)  ├ IoT Scheduler
                           └ Arctic (perma)    └ DND Priority

IoT Devices ──MQTT/AMQP──▶ Ingestion ──▶ Priority Queue ──▶ Scheduler
(sensors, drones)          (validate,    (URLLC first,       (PRB allocation,
                            classify,     defence boost,      preemption,
                            route)        shed mMTC)          congestion alert)

TN Signal ──▶ ┐
              ├──▶ Failover Engine ──▶ switch_to_tn / switch_to_leo / buffer / hard_fallback
LEO Signal ──▶ ┘   (anti-flapping,     (Rule R-6: buffer 30s → reroute → terrestrial fallback)
                     10s guard)

For developers: Clone the repo, run pip install numpy pytest fastapi httpx uvicorn, then python -m pytest tests/ -v to see all 332 tests pass. Every module is independently testable. No API keys, no accounts, no external services needed except api.weather.gc.ca (free).

Get Started

Run Tests

git clone https://github.com/KachaJugaad/TelcoEdge.git
cd TelcoEdge
pip install numpy pytest fastapi httpx uvicorn
python -m pytest tests/ -v

Run Demo

# Docker demo (Grafana + live weather)
docker compose -f deployment/docker-compose.demo.yml up
# Opens: localhost:3000

# Live map (no Docker needed)
uvicorn src.ran_intel.app:app --port 8080
# Opens: localhost:8080

Simulation vs Real World

What works now (simulation)	What needs real hardware
Weather data — live from Environment Canada	Already real — no simulation needed
Channel models — 4 terrains, 4000 runs	Field measurements to validate models
Policy decisions — all 6 policies tested	Real Near-RT RIC (OSC or commercial)
Latency budget — all hops pass in sim	Real gNB + air interface for Uu hop
NTN handover — F1 >= 0.80 in sim	Telesat LEO signal (API access pending)
IoT scheduling — preemption tested	Real MQTT broker + IoT sensors
PROTECTED-B compliance — rules enforced	DND security audit for production

What this means: The software is complete and tested. To move from simulation to production, you need a telco partner (TELUS, Rogers, Bell) with a real cell tower and an O-RAN RIC. The weather data is already real — it comes live from the Government of Canada right now.

Canadian Data Sovereignty

Weather data source	Government of Canada (MSC GeoMet) — free, anonymous, no key
Data stays in Canada?	Yes — all processing is local, no US/EU endpoints
Defence data classification	PROTECTED-B enforced — encryption in transit, Canadian-only
Every API call logged?	Yes — full audit trail in data/api_logs/
Security models?	STRIDE threat models for all API boundaries

Standards Used

Standard	What it is	How we use it
3GPP TR 38.901	Radio signal propagation model	RMa path loss for all 4 terrains
ITU-R P.838-3	Rain attenuation model	How much rain weakens signal at 3.5 GHz
ITU-R P.833-9	Vegetation loss model	Boreal forest foliage blockage
ITU-R P.526	Diffraction model	Rocky mountain ridge obstruction
O-RAN E2SM-RC v1.03	Radio control interface	Send MCS/beam changes to gNB
O-RAN E2SM-KPM v3.0	Performance metrics	Receive throughput/RSRP from gNB
3GPP TR 38.821	Satellite (NTN) spec	LEO handover prediction
3GPP TS 38.214	MCS index table	Modulation settings 0-28
Bill C-26	Canadian cyber security law	PROTECTED-B compliance

Who Can Use This

If you are...	Start here
A telecom engineer	Read `src/policies/` — 6 O-RAN policy classes, all tested. Fork and adapt for your RIC.
A researcher	Read `src/channel_plugins/` — 4 Canadian terrain models with 3GPP validation. Use for your own simulations.
A student	Run `python -m pytest tests/ -v` — see how a real telecom system is tested. Read the architecture doc.
A defence engineer	Read `src/defence/` and `docs/security/` — STRIDE models, PROTECTED-B checks, DND priority queue.
An investor/partner	Read `docs/demo_script.md` — 15-minute demo. Run `docker compose up` to see it live.
A hiring manager	332 tests, 9 standards, 6 policies, 4 terrain models, zero credentials. All open source.